Cloud MCP servers (Gmail, Google Calendar) auto-injected without user consent — no opt-out mechanism

Problem

Claude Code injects cloud-based MCP servers from claude.ai (specifically claude.ai Gmail and claude.ai Google Calendar) into the local CLI without any user opt-in. Every time Claude Code starts, it reports these servers need authentication — for services the user never installed, never agreed to, and does not want.

This is related to #20412 but focuses on the consent and privacy angle rather than the memory/OOM issue.

Why this matters

• No consent was given. I never installed, enabled, or agreed to Google/Gmail integration. These appeared in my local terminal uninvited.
• Privacy concern. Not every user trusts Google with their data. Having Google services surface in a local development tool — prompting for authentication on every startup — is unacceptable when the user never opted in.
• Cannot be removed. claude mcp remove does not work for these cloud-injected servers. The only workaround is editing a cached feature flag (tengu_claudeai_mcp_connectors) in ~/.claude.json, which may be overwritten on update.
• Annoying on every startup. The "needs authentication" warnings appear every time Claude Code launches, nagging the user to connect services they explicitly do not want.

Expected behavior

1. Cloud MCP servers should be opt-in only — never injected without explicit user action
2. Users should be able to permanently disable/remove cloud MCPs via claude mcp remove or a config setting
3. A local CLI tool should not push third-party service integrations (especially Google) without consent

Workaround

Edit ~/.claude.json and set:

"tengu_claudeai_mcp_connectors": false

inside cachedGrowthBookFeatures. This is fragile and may be overwritten.

Environment

• Claude Code (macOS)
• The servers appear as: claude.ai Gmail, claude.ai Google Calendar